EngineX-Cambricon/enginex-mlu370-any2any
9
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

init

Browse Source
This commit is contained in:
luopingyi
2025-10-09 16:47:16 +08:00
parent c8feb4deb5
commit e27e3f16bb
5248 changed files with 1778505 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

0
transformers/tests/models/superpoint/__init__.py Normal file
View File

181
transformers/tests/models/superpoint/test_image_processing_superpoint.py Normal file
View File

@@ -0,0 +1,181 @@
# Copyright 2024 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import unittest
import numpy as np
from transformers.testing_utils import require_torch, require_vision
from transformers.utils import is_torch_available, is_torchvision_available, is_vision_available
from ...test_image_processing_common import ImageProcessingTestMixin, prepare_image_inputs
if is_torch_available():
import torch
from transformers.models.superpoint.modeling_superpoint import SuperPointKeypointDescriptionOutput
if is_vision_available():
from transformers import SuperPointImageProcessor
if is_torchvision_available():
from transformers import SuperPointImageProcessorFast
class SuperPointImageProcessingTester:
def __init__(
self,
parent,
batch_size=7,
num_channels=3,
image_size=18,
min_resolution=30,
max_resolution=400,
do_resize=True,
size=None,
do_grayscale=True,
):
size = size if size is not None else {"height": 480, "width": 640}
self.parent = parent
self.batch_size = batch_size
self.num_channels = num_channels
self.image_size = image_size
self.min_resolution = min_resolution
self.max_resolution = max_resolution
self.do_resize = do_resize
self.size = size
self.do_grayscale = do_grayscale
def prepare_image_processor_dict(self):
return {
"do_resize": self.do_resize,
"size": self.size,
"do_grayscale": self.do_grayscale,
}
def expected_output_image_shape(self, images):
return self.num_channels, self.size["height"], self.size["width"]
def prepare_image_inputs(self, equal_resolution=False, numpify=False, torchify=False):
return prepare_image_inputs(
batch_size=self.batch_size,
num_channels=self.num_channels,
min_resolution=self.min_resolution,
max_resolution=self.max_resolution,
equal_resolution=equal_resolution,
numpify=numpify,
torchify=torchify,
)
def prepare_keypoint_detection_output(self, pixel_values):
max_number_keypoints = 50
batch_size = len(pixel_values)
mask = torch.zeros((batch_size, max_number_keypoints))
keypoints = torch.zeros((batch_size, max_number_keypoints, 2))
scores = torch.zeros((batch_size, max_number_keypoints))
descriptors = torch.zeros((batch_size, max_number_keypoints, 16))
for i in range(batch_size):
random_number_keypoints = np.random.randint(0, max_number_keypoints)
mask[i, :random_number_keypoints] = 1
keypoints[i, :random_number_keypoints] = torch.rand((random_number_keypoints, 2))
scores[i, :random_number_keypoints] = torch.rand((random_number_keypoints,))
descriptors[i, :random_number_keypoints] = torch.rand((random_number_keypoints, 16))
return SuperPointKeypointDescriptionOutput(
loss=None, keypoints=keypoints, scores=scores, descriptors=descriptors, mask=mask, hidden_states=None
)
@require_torch
@require_vision
class SuperPointImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
image_processing_class = SuperPointImageProcessor if is_vision_available() else None
fast_image_processing_class = SuperPointImageProcessorFast if is_torchvision_available() else None
def setUp(self) -> None:
super().setUp()
self.image_processor_tester = SuperPointImageProcessingTester(self)
@property
def image_processor_dict(self):
return self.image_processor_tester.prepare_image_processor_dict()
def test_image_processing(self):
for image_processing_class in self.image_processor_list:
image_processing = image_processing_class(**self.image_processor_dict)
self.assertTrue(hasattr(image_processing, "do_resize"))
self.assertTrue(hasattr(image_processing, "size"))
self.assertTrue(hasattr(image_processing, "do_rescale"))
self.assertTrue(hasattr(image_processing, "rescale_factor"))
self.assertTrue(hasattr(image_processing, "do_grayscale"))
def test_image_processor_from_dict_with_kwargs(self):
for image_processing_class in self.image_processor_list:
image_processor = image_processing_class.from_dict(self.image_processor_dict)
self.assertEqual(image_processor.size, {"height": 480, "width": 640})
image_processor = self.image_processing_class.from_dict(
self.image_processor_dict, size={"height": 42, "width": 42}
)
self.assertEqual(image_processor.size, {"height": 42, "width": 42})
@unittest.skip(reason="SuperPointImageProcessor is always supposed to return a grayscaled image")
def test_call_numpy_4_channels(self):
pass
def test_input_image_properly_converted_to_grayscale(self):
for image_processing_class in self.image_processor_list:
image_processor = image_processing_class.from_dict(self.image_processor_dict)
image_inputs = self.image_processor_tester.prepare_image_inputs()
pre_processed_images = image_processor.preprocess(image_inputs)
for image in pre_processed_images["pixel_values"]:
if isinstance(image, torch.Tensor):
self.assertTrue(
torch.all(image[0, ...] == image[1, ...]).item()
and torch.all(image[1, ...] == image[2, ...]).item()
)
else:
self.assertTrue(np.all(image[0, ...] == image[1, ...]) and np.all(image[1, ...] == image[2, ...]))
@require_torch
def test_post_processing_keypoint_detection(self):
def check_post_processed_output(post_processed_output, image_size):
for post_processed_output, image_size in zip(post_processed_output, image_size):
self.assertTrue("keypoints" in post_processed_output)
self.assertTrue("descriptors" in post_processed_output)
self.assertTrue("scores" in post_processed_output)
keypoints = post_processed_output["keypoints"]
all_below_image_size = torch.all(keypoints[:, 0] <= image_size[1]) and torch.all(
keypoints[:, 1] <= image_size[0]
)
all_above_zero = torch.all(keypoints[:, 0] >= 0) and torch.all(keypoints[:, 1] >= 0)
self.assertTrue(all_below_image_size)
self.assertTrue(all_above_zero)
for image_processing_class in self.image_processor_list:
image_processor = image_processing_class.from_dict(self.image_processor_dict)
image_inputs = self.image_processor_tester.prepare_image_inputs()
pre_processed_images = image_processor.preprocess(image_inputs, return_tensors="pt")
outputs = self.image_processor_tester.prepare_keypoint_detection_output(**pre_processed_images)
tuple_image_sizes = [(image.size[0], image.size[1]) for image in image_inputs]
tuple_post_processed_outputs = image_processor.post_process_keypoint_detection(outputs, tuple_image_sizes)
check_post_processed_output(tuple_post_processed_outputs, tuple_image_sizes)
tensor_image_sizes = torch.tensor([image.size for image in image_inputs]).flip(1)
tensor_post_processed_outputs = image_processor.post_process_keypoint_detection(
outputs, tensor_image_sizes
)
check_post_processed_output(tensor_post_processed_outputs, tensor_image_sizes)

312
transformers/tests/models/superpoint/test_modeling_superpoint.py Normal file
View File

@@ -0,0 +1,312 @@
# Copyright 2024 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import inspect
import unittest
from functools import cached_property
from transformers.models.superpoint.configuration_superpoint import SuperPointConfig
from transformers.testing_utils import is_flaky, require_torch, require_vision, slow, torch_device
from transformers.utils import is_torch_available, is_vision_available
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, floats_tensor
if is_torch_available():
import torch
from transformers import (
SuperPointForKeypointDetection,
)
if is_vision_available():
from PIL import Image
from transformers import AutoImageProcessor
class SuperPointModelTester:
def __init__(
self,
parent,
batch_size=3,
image_width=80,
image_height=60,
encoder_hidden_sizes: list[int] = [32, 32, 64, 64],
decoder_hidden_size: int = 128,
keypoint_decoder_dim: int = 65,
descriptor_decoder_dim: int = 128,
keypoint_threshold: float = 0.005,
max_keypoints: int = -1,
nms_radius: int = 4,
border_removal_distance: int = 4,
):
self.parent = parent
self.batch_size = batch_size
self.image_width = image_width
self.image_height = image_height
self.encoder_hidden_sizes = encoder_hidden_sizes
self.decoder_hidden_size = decoder_hidden_size
self.keypoint_decoder_dim = keypoint_decoder_dim
self.descriptor_decoder_dim = descriptor_decoder_dim
self.keypoint_threshold = keypoint_threshold
self.max_keypoints = max_keypoints
self.nms_radius = nms_radius
self.border_removal_distance = border_removal_distance
def prepare_config_and_inputs(self):
# SuperPoint expects a grayscale image as input
pixel_values = floats_tensor([self.batch_size, 3, self.image_height, self.image_width])
config = self.get_config()
return config, pixel_values
def get_config(self):
return SuperPointConfig(
encoder_hidden_sizes=self.encoder_hidden_sizes,
decoder_hidden_size=self.decoder_hidden_size,
keypoint_decoder_dim=self.keypoint_decoder_dim,
descriptor_decoder_dim=self.descriptor_decoder_dim,
keypoint_threshold=self.keypoint_threshold,
max_keypoints=self.max_keypoints,
nms_radius=self.nms_radius,
border_removal_distance=self.border_removal_distance,
)
def create_and_check_keypoint_detection(self, config, pixel_values):
model = SuperPointForKeypointDetection(config=config)
model.to(torch_device)
model.eval()
result = model(pixel_values)
self.parent.assertEqual(result.keypoints.shape[0], self.batch_size)
self.parent.assertEqual(result.keypoints.shape[-1], 2)
result = model(pixel_values, output_hidden_states=True)
self.parent.assertEqual(
result.hidden_states[-1].shape,
(
self.batch_size,
self.encoder_hidden_sizes[-1],
self.image_height // 8,
self.image_width // 8,
),
)
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
config, pixel_values = config_and_inputs
inputs_dict = {"pixel_values": pixel_values}
return config, inputs_dict
@require_torch
class SuperPointModelTest(ModelTesterMixin, unittest.TestCase):
all_model_classes = (SuperPointForKeypointDetection,) if is_torch_available() else ()
fx_compatible = False
test_pruning = False
test_resize_embeddings = False
test_head_masking = False
has_attentions = False
from_pretrained_id = "magic-leap-community/superpoint"
def setUp(self):
self.model_tester = SuperPointModelTester(self)
self.config_tester = ConfigTester(
self,
config_class=SuperPointConfig,
has_text_modality=False,
hidden_size=37,
common_properties=["encoder_hidden_sizes", "decoder_hidden_size"],
)
def test_config(self):
self.config_tester.run_common_tests()
@is_flaky(description="The `indices` computed with `topk()` in `top_k_keypoints` is not stable.")
def test_batching_equivalence(self):
super().test_batching_equivalence()
@unittest.skip(reason="SuperPointForKeypointDetection does not use inputs_embeds")
def test_inputs_embeds(self):
pass
@unittest.skip(reason="SuperPointForKeypointDetection does not support input and output embeddings")
def test_model_get_set_embeddings(self):
pass
@unittest.skip(reason="SuperPointForKeypointDetection does not use feedforward chunking")
def test_feed_forward_chunking(self):
pass
@unittest.skip(reason="SuperPointForKeypointDetection does not support training")
def test_training(self):
pass
@unittest.skip(reason="SuperPointForKeypointDetection does not support training")
def test_training_gradient_checkpointing(self):
pass
@unittest.skip(reason="SuperPointForKeypointDetection does not support training")
def test_training_gradient_checkpointing_use_reentrant(self):
pass
@unittest.skip(reason="SuperPointForKeypointDetection does not support training")
def test_training_gradient_checkpointing_use_reentrant_false(self):
pass
@unittest.skip(reason="SuperPoint does not output any loss term in the forward pass")
def test_retain_grad_hidden_states_attentions(self):
pass
def test_keypoint_detection(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_keypoint_detection(*config_and_inputs)
def test_forward_signature(self):
config, _ = self.model_tester.prepare_config_and_inputs()
for model_class in self.all_model_classes:
model = model_class(config)
signature = inspect.signature(model.forward)
# signature.parameters is an OrderedDict => so arg_names order is deterministic
arg_names = [*signature.parameters.keys()]
expected_arg_names = ["pixel_values"]
self.assertListEqual(arg_names[:1], expected_arg_names)
def test_hidden_states_output(self):
def check_hidden_states_output(inputs_dict, config, model_class):
model = model_class(config)
model.to(torch_device)
model.eval()
with torch.no_grad():
outputs = model(**self._prepare_for_class(inputs_dict, model_class))
hidden_states = outputs.hidden_states
# SuperPoint's feature maps are of shape (batch_size, num_channels, width, height)
for i, conv_layer_size in enumerate(self.model_tester.encoder_hidden_sizes[:-1]):
self.assertListEqual(
list(hidden_states[i].shape[-3:]),
[
conv_layer_size,
self.model_tester.image_height // (2 ** (i + 1)),
self.model_tester.image_width // (2 ** (i + 1)),
],
)
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
inputs_dict["output_hidden_states"] = True
check_hidden_states_output(inputs_dict, config, model_class)
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
config.output_hidden_states = True
check_hidden_states_output(inputs_dict, config, model_class)
@slow
def test_model_from_pretrained(self):
model = SuperPointForKeypointDetection.from_pretrained(self.from_pretrained_id)
self.assertIsNotNone(model)
def test_forward_labels_should_be_none(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
model = model_class(config)
model.to(torch_device)
model.eval()
with torch.no_grad():
model_inputs = self._prepare_for_class(inputs_dict, model_class)
# Provide an arbitrary sized Tensor as labels to model inputs
model_inputs["labels"] = torch.rand((128, 128))
with self.assertRaises(ValueError) as cm:
model(**model_inputs)
self.assertEqual(ValueError, cm.exception.__class__)
def prepare_imgs():
image1 = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png")
image2 = Image.open("./tests/fixtures/tests_samples/COCO/000000004016.png")
return [image1, image2]
@require_torch
@require_vision
class SuperPointModelIntegrationTest(unittest.TestCase):
@cached_property
def default_image_processor(self):
return AutoImageProcessor.from_pretrained("magic-leap-community/superpoint") if is_vision_available() else None
@slow
def test_inference(self):
model = SuperPointForKeypointDetection.from_pretrained("magic-leap-community/superpoint").to(torch_device)
preprocessor = self.default_image_processor
images = prepare_imgs()
inputs = preprocessor(images=images, return_tensors="pt").to(torch_device)
with torch.no_grad():
outputs = model(**inputs)
expected_number_keypoints_image0 = 568
expected_number_keypoints_image1 = 830
expected_max_number_keypoints = max(expected_number_keypoints_image0, expected_number_keypoints_image1)
expected_keypoints_shape = torch.Size((len(images), expected_max_number_keypoints, 2))
expected_scores_shape = torch.Size(
(
len(images),
expected_max_number_keypoints,
)
)
expected_descriptors_shape = torch.Size((len(images), expected_max_number_keypoints, 256))
# Check output shapes
self.assertEqual(outputs.keypoints.shape, expected_keypoints_shape)
self.assertEqual(outputs.scores.shape, expected_scores_shape)
self.assertEqual(outputs.descriptors.shape, expected_descriptors_shape)
expected_keypoints_image0_values = torch.tensor([[0.75, 0.0188], [0.7719, 0.0188], [0.7641, 0.0333]]).to(
torch_device
)
expected_scores_image0_values = torch.tensor(
[0.0064, 0.0139, 0.0591, 0.0727, 0.5170, 0.0175, 0.1526, 0.2057, 0.0335]
).to(torch_device)
expected_descriptors_image0_value = torch.tensor(-0.1095).to(torch_device)
predicted_keypoints_image0_values = outputs.keypoints[0, :3]
predicted_scores_image0_values = outputs.scores[0, :9]
predicted_descriptors_image0_value = outputs.descriptors[0, 0, 0]
# Check output values
self.assertTrue(
torch.allclose(
predicted_keypoints_image0_values,
expected_keypoints_image0_values,
atol=1e-4,
)
)
torch.testing.assert_close(predicted_scores_image0_values, expected_scores_image0_values, rtol=1e-4, atol=1e-4)
self.assertTrue(
torch.allclose(
predicted_descriptors_image0_value,
expected_descriptors_image0_value,
atol=1e-4,
)
)
# Check mask values
self.assertTrue(outputs.mask[0, expected_number_keypoints_image0 - 1].item() == 1)
self.assertTrue(outputs.mask[0, expected_number_keypoints_image0].item() == 0)
self.assertTrue(torch.all(outputs.mask[0, : expected_number_keypoints_image0 - 1]))
self.assertTrue(torch.all(torch.logical_not(outputs.mask[0, expected_number_keypoints_image0:])))
self.assertTrue(torch.all(outputs.mask[1]))